A study has found that relying on artificial intelligence (AI) assistants such as ChatGPT may reduce brain activity during writing tasks and contribute to what researchers call a lasting “cognitive debt.”
The neuroimaging study, led by Nataliya Kosmyna and published on arXiv, used electroencephalogram (EEG) recordings to examine the neural, behavioral, and memory effects of using large language models (LLMs) for analytical writing tasks.
To conduct the study, researchers divided participants into three groups, each using a different approach to writing.
The first group, called the LLM group, used ChatGPT. The second group, the Search Engine group, relied on traditional web searches without AI assistance. The third group, called the “Brain-only” group, completed their writing tasks entirely without external assistance.
EEG analysis revealed that brain connectivity decreased as the level of external assistance increased.
Participants in the LLM group exhibited the weakest neural connectivity, with reductions of up to 55% compared with the unassisted group. The researchers linked these patterns to neural processes associated with internal monitoring and critical thinking.
The Search Engine group showed intermediate levels of engagement, with neural connectivity 34% to 48% lower than that of the unassisted writers.
Meanwhile, the “Brain-only” group exhibited the strongest and most extensive neural networks across all frequency bands, reflecting greater engagement in cognitive processing, working memory, and active idea generation.
EXAMINING “COGNITIVE DEBT”
In a fourth session, researchers examined how “cognitive debt” might emerge when participants switched between AI-assisted and unassisted writing.
In the LLM-to-brain condition, habitual ChatGPT users were asked to write without AI tools. Their brains continued to show reduced neural engagement, which the researchers interpreted as a potential sign of persistent “cognitive debt” affecting active critical thinking.
In contrast, participants who initially wrote without assistance were given access to ChatGPT in the brain-to-LLM condition. Their brains showed a substantial increase in neural activity, suggesting greater cognitive effort as they evaluated, filtered, and integrated AI-generated suggestions.
Researchers concluded that while AI tools can make essay writing faster and more efficient, excessive reliance on them may undermine foundational learning and critical-thinking skills over time.
“It becomes crucial to understand the full spectrum of cognitive consequences associated with LLM integration in educational and informational contexts,” they said. “While these tools offer unprecedented opportunities for enhancing learning and information access, their potential impact on cognitive development, critical thinking, and intellectual independence demands a very careful consideration and continued research.”
The researchers further emphasized the need for educators and professionals to better understand the effects of generative AI on learning and education.
